Redaction Gone Wrong: Why Black Boxes Keep Leaking Secrets
Every few years the same news story repeats with different names: a court filing, a government report, or a corporate disclosure is published with black bars over the sensitive parts — and within hours, someone on the internet has read everything under them. High-profile legal cases have leaked sealed names this way; military reports have exposed classified passages. The embarrassing constant across all of these incidents is that the black boxes worked perfectly on the eyes and not at all on the file.
The failure has a simple mechanical cause. A PDF page is a stack of drawing instructions: the text is one instruction, and a black rectangle drawn on top of it is simply another instruction, later in the stack. The words are still in the file — selectable, searchable, copyable. Select the area, copy, paste into a text editor, and the 'redacted' content reappears. The same applies to related half-measures: white text on white background, images cropped only visually (the full image often remains in the file), and highlighter annotations set to black.
True redaction is not covering — it is removal. The text objects must be deleted from the content stream, the region of any underlying image must be actually blanked, and the visible black box is merely a cosmetic marker of where something used to be. After genuine redaction, copy-paste yields nothing because there is nothing left to copy. This is precisely what PdfWill's redact tool does, and because it runs in your browser, the unredacted original — the most sensitive version of your document — never travels to any server.
Two more leak channels deserve attention even after proper redaction. Metadata can carry author names, file paths, and revision comments; and if the document is a scan, remember that OCR can resurrect text from images — meaning the image region itself must be blanked, not just any text layer above it. When the stakes are high, a belt-and-braces verification is simple: run the redacted file through OCR or select-all-and-copy, and confirm the sensitive strings are truly gone.
The lesson from two decades of redaction failures is not that PDFs are treacherous; it is that visual inspection is the wrong test for an information problem. A document is redacted when the data is gone from the file — not when your screen shows a black box. Tools that do it properly exist and are free; the only unrecoverable mistake is publishing first and checking second.